README.md

metadata

language:
  - fr
dataset_info:
  features:
    - name: context
      dtype: string
    - name: question
      dtype: string
    - name: label
      dtype: int64
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 292674100
      num_examples: 234708
    - name: test
      num_bytes: 49719642
      num_examples: 33610
  download_size: 57567043
  dataset_size: 342393742
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

This is a dataset for training the cross-encoder of our RAG system. It is a combination of the PIAF, FQuAD, SQuAD-French, and pandora-s-fr datasets.